|
(11) | EP 2 318 926 B1 |
| (12) | EUROPEAN PATENT SPECIFICATION |
|
|
| (54) |
METHOD AND SYSTEM FOR LOW-OVERHEAD DATA TRANSFER VERFAHREN UND SYSTEM FÜR DATENTRANSFER MIT GERINGEM OVERHEAD PROCÉDÉ ET SYSTÈME POUR UN TRANSFERT DE DONNÉES À FAIBLE SURDÉBIT |
|
|
|||||||||||||||||||||||||||||||
| Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention). |
BACKGROUND
[0001] Conventionally, in the computer-related arts, a network is an arrangement of physical computer systems configured to communicate with each other. In some cases, the physical computer systems include virtual machines, which may also be configured to interact with the network (i.e., communicate with other physical computers and/or virtual machines in the network). Many different types of networks exist, and a network may be classified based on various aspects of the network, such as scale, connection method, functional relationship of computer systems in the network, and/or network topology.
[0002] Regarding connection methods, a network may be broadly categorized as wired (using a tangible connection medium such as Ethernet cables) or wireless (using an intangible connection medium such as radio waves). Different connection methods may also be combined in a single network. For example, a wired network may be extended to allow devices to connect to the network wirelessly. However, core network components such as routers, switches, and servers are generally connected using physical wires. Ethernet is defined within the Institute of Electrical and Electronics Engineers (IEEE) 802.3 standards, which are supervised by the IEEE 802.3 Working Group.
[0003] To create a wired network, computer systems must be physically connected to each other. That is, the ends of physical wires (for example, Ethernet cables) must be physically connected to network interface cards in the computer systems forming the network. To reconfigure the network (for example, to replace a server or change the network topology), one or more of the physical wires must be disconnected from a computer system and connected to a different computer system.
[0004] Further, when transferring data between computer systems in a network, one or more network protocols are typically used to help ensure the data are transferred successfully. For example, network protocols may use checksums, small data packets, acknowledgments, and other data integrity features to help avoid data loss or corruption during the data transfer. The amount of data integrity features required in the network protocol(s) generally depends on the type of data being transferred and the quality of the connection(s) between the computer systems.
[0005] An article by D Goldenberg et al, entitled "Zero Copy Sockets Direct Protocol over InfiniBand - Preliminary Implementation and Performance Analysis", published in the Proceedings of the 13th Symposium on High Performance Interconnects, 17-19 August 2005 (ISBN:0-7695-2449-4) describes the Sockets Direct Protocol (SDP), which is a byte-stream transport protocol implementing the TCP SOCK_STREAM semantics utilizing transport offloading capabilities of the InfiniBand fabric. SDP supports a Zero-Copy (ZCopy) operation mode, using the InfiniBand RDMA capability to transfer data directly between application buffers. Alternatively, in a Buffer Copy (BCopy) mode, data is copied to and from transport buffers. The paper presents architectural aspects of the SDP protocol, the ZCopy implementation, and a preliminary performance evaluation.
SUMMARY
[0007] In general, in one aspect, the invention relates to a method for low-overhead data transfer. The method includes initiating, by a first application, a Transmission Communication Protocol (TCP) connection with a second application, wherein the first application is configured to execute in a first virtual machine on a first computer, the second application is configured to execute in a second virtual machine on a second computer, and the first computer and the second computer are located on a chassis and communicate over a chassis interconnect, establishing, in response to the initiation, the TCP connection between the first application and the second application, determining that the first computer and second computer are located on the chassis and, responsive to determining that the first computer and second computer are located on the chassis, providing, by the first application, pre-post buffer information to the second application, wherein the pre-post buffer information corresponds to a location in a physical memory of the first computer and wherein the location in physical memory corresponds to a virtual memory address of the first application, and transferring data, by the second application, to the first application using the pre-post buffer information, wherein transferring the data comprises writing the data directly into the location in the physical memory of the first computer.
[0008] In general, in one aspect, the invention relates to a program product, for example embodied on a computer readable medium, comprising a plurality of executable instructions for low-overhead data transfer, wherein the plurality of executable instructions comprises instructions to carry out such a method.
[0009] In general, in one aspect, the invention relates to a system. The system includes a chassis interconnect and a first application is executing on a first computer in a first virtual machine and a second application is executing on a second computer in a second virtual machine, wherein the first computer and the second computer are located on a chassis and communicate over the chassis interconnect, wherein the first application is configured to initiate a Transmission Communication Protocol (TCP) connection with the second application, wherein, in response to the initiation, the TCP connection is established between the first application and the second application, wherein the first application is configured to provide pre-post buffer information to the second application in response to a determination that the first application is executing on the same chassis as the second application, wherein the pre-post buffer information corresponds to a location in a physical memory of the first computer and wherein the location in physical memory corresponds to a virtual memory address of the first application, and wherein the second application transfers data to the first application using the pre-post buffer information, wherein transferring the data comprises writing the data directly into the location in the physical memory of the first computer.
[0010] Other aspects of the invention will be apparent from the following description and the appended claims.
BRIEF DESCRIPTION OF DRAWINGS
[0011]
Figure 1 shows a diagram of a blade chassis in accordance with one or more embodiments of the invention.
Figure 2 shows a diagram of a blade in accordance with one or more embodiments of the invention.
Figure 3 shows a diagram of a network express manager in accordance with one or more embodiments of the invention.
Figure 4 shows a diagram of a virtual machine in accordance with one or more embodiments of the invention.
Figure 5 shows a flowchart of a method for creating a virtual network path in accordance with one or more embodiments of the invention.
Figures 6A-6C show an example of creating virtual network paths in accordance with one or more embodiments of the invention.
Figures 7-8 show flowcharts of a method for low-overhead data transfer in accordance with one or more embodiments of the invention.
Figure 9 shows an example of low-overhead data transfer in accordance with one or more embodiments of the invention.
DETAILED DESCRIPTION
[0012] Specific embodiments of the invention will now be described in detail, by way of example only, with reference to the accompanying figures. Like elements in the various figures are denoted by like reference numerals for consistency.
[0013] In the following detailed description of embodiments of the invention, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to one of ordinary skill in the art that the invention may be practiced without these specific details. In other instances, well-known features have not been described in detail to avoid unnecessarily complicating the description.
[0014] In general, embodiments of the invention provide a method and system for low-overhead data transfer. More specifically, embodiments of the invention provide a method and system for enabling two applications executing on blades within a common blade chassis to communicate using low-overhead data transfer. Further, embodiments of the invention provide a method and system to enable two applications to participate in a zero-copy handshake and then proceed to communicate using low-overhead data transfer.
[0015] In one or more embodiments of the invention, the VNICs are connected to each other via a chassis interconnect. Specifically, the VNICs may be nodes of a virtual network path that includes a "virtual wire" used to transmit network traffic via the chassis interconnect. The concept of a virtual wire is discussed in detail below.
[0016] Figure 1 shows a diagram of a blade chassis (100) in accordance with one or more embodiments of the invention. The blade chassis (100) includes multiple blades (e.g., blade A (102), blade B (104)) communicatively coupled with a chassis interconnect (106). For example, the blade chassis (100) may be a Sun Blade 6048 Chassis by Sun Microsystems Inc., an IBM BladeCenter® chassis, an HP BladeSystem enclosure by Hewlett Packard Inc., or any other type of blade chassis. The blades may be of any type(s) compatible with the blade chassis (100). BladeCenter® is a registered trademark of International Business Machines, Inc. (IBM), headquartered in Armonk, New York.
[0017] In one or more embodiments of the invention, the blades are configured to communicate with each other via the chassis interconnect (106). Thus, the blade chassis (100) allows for communication between the blades without requiring traditional network wires (such as Ethernet cables) between the blades. For example, depending on the type of blade chassis (100), the chassis interconnect (106) may be a Peripheral Component Interface Express (PCI-E) backplane, and the blades may be configured to communicate with each other via PCI-E endpoints. Those skilled in the art will appreciate that other connection technologies may be used to connect the blades to the blade chassis.
[0018] Continuing with the discussion of Figure 1, to communicate with clients outside the blade chassis (100), the blades are configured to share a physical network interface (110). The physical network interface (110) includes one or more network ports (for example, Ethernet ports), and provides an interface between the blade chassis (100) and the network (i.e., interconnected computer systems external to the blade chassis (100)) to which the blade chassis (100) is connected. The blade chassis (100) may be connected to multiple networks, for example using multiple network ports.
[0019] In one or more embodiments, the physical network interface (110) is managed by a network express manager (108). Specifically, the network express manager (108) is configured to manage access by the blades to the physical network interface (110). The network express manager (108) may also be configured to manage internal communications between the blades themselves, in a manner discussed in detail below. The network express manager (108) may be any combination of hardware, software, and/or firmware including executable logic for managing network traffic.
[0020] Figure 2 shows a diagram of a blade (200) in accordance with one or more embodiments of the invention. "Blade" is a term of art referring to a computer system located within a blade chassis (for example, the blade chassis (100) of Figure 1). Blades typically include fewer components than stand-alone computer systems or conventional servers. In one or more embodiments of the invention, fully featured stand-alone computer systems or conventional servers may also be used instead of (or in combination with) the blades. Generally, blades in a blade chassis each include one or more processors and associated memory (e.g., RAM, ROM, etc.). Blades may also include storage devices (for example, hard drives and/or optical drives) and numerous other elements and functionalities typical of today's computer systems (not shown), such as a keyboard, a mouse, and/or output means such as a monitor. One or more of the aforementioned components may be shared by multiple blades located in the blade chassis. For example, multiple blades may share a single output device.
[0021] Continuing with discussion of Figure 2, the blade (200) includes a host operating system (not shown) configured to execute one or more virtual machines (e.g., virtual machine C (202), virtual machine D (204)). Broadly speaking, the virtual machines are distinct operating environments configured to inherit underlying functionality of the host operating system via an abstraction layer. In one or more embodiments of the invention, each virtual machine includes a separate instance of an operating system (e.g., operating system instance C (206), operating system instance D (208)). For example, the Xen® virtualization project allows for multiple guest operating systems executing in a host operating system. Xen® is a trademark overseen by the Xen Project Advisory Board. In one or more embodiments of the invention, the host operating system supports virtual execution environments (not shown). An example of virtual execution environment is a Solaris™ Container. In such cases, the Solaris™ Container may execute in the host operating system, which may be a Solaris™ operating system. Solaris™ is a trademark of Sun Microsystems, Inc. In one or more embodiments of the invention, the host operating system may include both virtual machines and virtual execution environments.
[0022] Many different types of virtual machines and virtual execution environment exist. Further, the virtual machines may include many different types of functionality, such as a switch, a router, a firewall, a load balancer, an application server, any other type of network-enabled service, or any combination thereof.
[0023] In one or more embodiments of the invention, the virtual machines and/or virtual execution environments inherit network connectivity from the host operating system via VNICs (e.g., VNIC C (210), VNIC D (212)). To the virtual machines and the virtual execution environments, the VNICs appear as physical NICs. In one or more embodiments of the invention, the use of VNICs allows an arbitrary number of virtual machines and/or virtual execution environments to share the blade's (200) networking functionality. Further, in one or more embodiments of the invention, each virtual machine and/or virtual execution environment may be associated with an arbitrary number of VNICs, thereby providing increased flexibility in the types of networking functionality available to the virtual machines and/or virtual execution environments. For example, a virtual machine may use one VNIC for incoming network traffic, and another VNIC for outgoing network traffic.
[0024] VNICs in accordance with one or more embodiments of the invention are described in detail in commonly owned U.S. Patent Application Serial No. 11/489,942, entitled "Multiple Virtual Network Stack Instances using Virtual Network Interface Cards," in the names of Nicolas G. Droux, Erik Nordmark, and Sunay Tripathi, the contents of which are hereby incorporated by reference in their entirety. VNICs in accordance with one or more embodiments of the invention also are described in detail in commonly owned U.S. Patent Application Serial No. 11/480,000, entitled "Method and System for Controlling Virtual Machine Bandwidth" in the names of Sunay Tripathi, Tim P. Marsland, and Nicolas G. Droux the contents of which are hereby incorporated by reference in their entirety.
[0025] As discussed above, each blade's networking functionality (and, by extension, networking functionality inherited by the VNICs) includes access to a shared physical network interface and communication with other blades via the chassis interconnect. Figure 3 shows a diagram of a network express manager (300) in accordance with one or more embodiments of the invention. The network express manager (300) is configured to route network traffic traveling to and from VNICs located in the blades. Specifically, the network express manager (300) includes a virtual switching table (302), which includes a mapping of VNIC identifiers (304) to VNIC locations (306) in the chassis interconnect. In one or more embodiments, the VNIC identifiers (304) are Internet Protocol (IP) addresses, and the VNIC locations (306) are PCI-E endpoints associated with the blades (e.g., if the chassis interconnect is a PCI-E backplane). Alternatively, another routing scheme may be used.
[0026] In one or more embodiments, the network express manager (300) is configured to receive network traffic via the physical network interface and route the network traffic to the appropriate location (i.e., where the VNIC is located) using the virtual switching table (302). Further, the network express manager (300) may be configured to route network traffic between different VNICs located in the blade chassis. In one or more embodiments of the invention, using the virtual switching table (302) in this manner facilitates the creation of a virtual network path, which includes virtual wires. Thus, using the virtual switching table (302), virtual machines located in different blades may be interconnected to form an arbitrary virtual network topology, where the VNICs associated with each virtual machine do not need to know the physical locations of other VNICs. Further, if a virtual machine is migrated from one blade to another, the virtual network topology may be preserved by updating the virtual switching table (302) to reflect the corresponding VNIC's new physical location (for example, a different PCI-E endpoint).
[0027] In some cases, network traffic from one VNIC may be destined for a VNIC located in the same blade, but associated with a different virtual machine. In one or more embodiments of the invention, a virtual switch may be used to route the network traffic between the VNICs independent of the blade chassis. Virtual switches in accordance with one or more embodiments of the invention are discussed in detail in commonly owned U.S. Patent Application Serial No. 11/480,261, entitled "Virtual Switch," in the names of Nicolas G. Droux, Sunay Tripathi, and Erik Nordmark, the contents of which are hereby incorporated by reference in their entirety.
[0028] For example, Figure 4 shows a diagram of a virtual switch (400) in accordance with one or more embodiments of the invention. The virtual switch (400) provides connectivity between VNIC X (406) associated with virtual machine X (402) and VNIC Y (408) associated with virtual machine Y (404). In one or more embodiments, the virtual switch (400) is managed by a host operating system (410) within which virtual machine X (402) and virtual machine Y (404) are located. Specifically, the host operating system (410) may be configured to identify network traffic targeted at a VNIC in the same blade, and route the traffic to the VNIC using the virtual switch (400). In one or more embodiments of the invention, the virtual switch (400) may reduce utilization of the blade chassis and the network express manager by avoiding unnecessary round-trip network traffic.
[0029] Figure 5 shows a flowchart of a method for creating a virtual network path in accordance with one or more embodiments of the invention. In one or more embodiments of the invention, one or more of the steps shown in Figure 5 may be omitted, repeated, and/or performed in a different order. Accordingly, embodiments of the invention should not be considered limited to the specific arrangement of steps shown in Figure 5.
[0030] In one or more embodiments of the invention, in Step 502, VNICs are instantiated for multiple virtual machines. The virtual machines are located in blades, as discussed above. Further, the virtual machines may each be associated with one or more VNICs. In one or more embodiments of the invention, instantiating a VNIC involves loading a VNIC object in memory and registering the VNIC object with a host operating system, i.e., an operating system that is hosting the virtual machine associated with the VNIC. Registering the VNIC object establishes an interface between the host operating system's networking functionality and the abstraction layer provided by the VNIC. Thereafter, when the host operating system receives network traffic addressed to the VNIC, the host operating system forwards the network traffic to the VNIC. Instantiation of VNICs in accordance with one or more embodiments of the invention is discussed in detail in U.S. Patent Application 11/489,942, incorporated by reference above.
[0031] As discussed above, a single blade may include multiple virtual machines configured to communicate with each other. In one or more embodiments of the invention, in Step 504, a virtual switch is instantiated to facilitate communication between the virtual machines. As noted above, the virtual switch allows communication between VNICs independent of the chassis interconnect. Instantiation of virtual switches in accordance with one or more embodiments of the invention is discussed in detail in U.S. Patent Application 11/480,261, incorporated by reference above.
[0032] In one or more embodiments of the invention, in Step 506, a virtual switching table is populated. As noted above, the virtual switching table may be located in a network express manager configured to manage network traffic flowing to and from the virtual machines. Populating the virtual switching table involves associating VNIC identifiers (for example, Internet Protocol and/or Media Access Control (MAC) addresses) with VNIC locations (for example, PCI-E endpoints). In one or more embodiments of the invention, the virtual switching table is populated in response to a user command issued via a control operating system, i.e., an operating system that includes functionality to control the network express manager.
[0033] In one or more embodiments of the invention, VNICs include settings for controlling the processing of network packets. In one or more embodiments of the invention, in Step 508, settings are assigned to the VNICs according to a networking policy. Many different types of networking policies may be enforced using settings in the VNICs. For example, a setting may be used to provision a particular portion of a blade's available bandwidth to one or more VNICs. As another example, a setting may be used to restrict use of a VNIC to a particular type of network traffic, such as Voice over IP (VoIP) or Transmission Control Protocol/IP (TCP/IP). Further, settings for multiple VNICs in a virtual network path may be identical. For example, VNICs in a virtual network path may be capped at the same bandwidth limit, thereby allowing for consistent data flow across the virtual network path. In one or more embodiments of the invention, a network express manager is configured to transmit the desired settings to the VNICs.
[0034] In one or more embodiments of the invention, once the VNICs are instantiated and the virtual switching table is populated, network traffic may be transmitted from a VNIC in one blade to a VNIC in another blade. The connection between the two VNICs may be thought of as a "virtual wire," because the arrangement obviates the need for traditional network wires such as Ethernet cables. A virtual wire functions similar to a physical wire in the sense that network traffic passing through one virtual wire is isolated from network traffic passing through another virtual wire, even though the network traffic may pass through the same blade (i.e., using the same virtual machine or different virtual machines located in the blade).
[0035] Further, a combination of two or more virtual wires may be thought of as a "virtual network path." Specifically, transmitting network traffic over the virtual network path involves routing the network traffic through a first virtual wire (Step 510) and then through a second virtual wire (Step 512). For example, when receiving network traffic from a client via the physical network interface, one virtual wire may be located between the physical network interface and a VNIC, and a second virtual wire may be located between the VNIC and another VNIC.
[0036] Figures 6A-6C show an example of creating virtual network paths in accordance with one or more embodiments of the invention. Specifically, Figure 6A shows a diagram of an actual topology (600) in accordance with one or more embodiments of the invention, Figure 6B shows how network traffic may be routed through the actual topology (600), and Figure 6C shows a virtual network topology (640) created by routing network traffic as shown in Figure 6B. Figures 6A-6C are provided as examples only, and should not be construed as limiting the scope of the invention.
[0037] Referring first to Figure 6A, the actual topology (600) includes multiple virtual machines. Specifically, the actual topology (600) includes a router (602), a firewall (604), application server M (606), and application server N (608), each executing in a separate virtual machine. The virtual machines are located in blades communicatively coupled with a chassis interconnect (622), and include networking functionality provided by the blades via VNICs (i.e., VNIC H (610), VNIC J (612), VNIC K (614), VNIC M (618), and VNIC N (620)). For ease of illustration, the blades themselves are not included in the diagram.
[0038] In one or more embodiments of the invention, the router (602), the firewall (604), application server M (606), and application server N (608) are each located in separate blades. Alternatively, as noted above, a blade may include multiple virtual machines. For example, the router (602) and the firewall (604) may be located in a single blade. Further, each virtual machine may be associated with a different number of VNICs than the number of VNICs shown in Figure 6A.
[0039] Continuing with discussion of Figure 6A, a network express manager (624) is configured to manage network traffic flowing to and from the virtual machines. Further, the network express manager (624) is configured to manage access to a physical network interface (626) used to communicate with client O (628) and client P (630). In Figure 6A, the virtual machines, VNICs, chassis interconnect (622), network express manager (624), and physical network interface (626) are all located within a chassis interconnect. Client O (628) and client P (630) are located in one or more networks (not shown) to which the chassis interconnect is connected.
[0040] Figure 6B shows how network traffic may be routed through the actual topology (600) in accordance with one or more embodiments of the invention. In one or more embodiments of the invention, the routing is performed by the network express manager (624) using a virtual switching table (634).
[0041] As discussed above, network traffic routed to and from the VNICs may be though of as flowing through a "virtual wire." For example, Figure 6B shows a virtual wire (632) located between application server M (606) and application server N (608). To use the virtual wire, application server M (606) transmits a network packet via VNIC M (618). The network packet is addressed to VNIC N (620) associated with application server N (608). The network express manager (624) receives the network packet via the chassis interconnect (622), inspects the network packet, and determines the target VNIC location using the virtual switching table (634). If the target VNIC location is not found in the virtual switching table (634), then the network packet may be dropped. In this example, the target VNIC location is the blade in which VNIC N (620) is located. The network express manager (624) routes the network packet to the target VNIC location, and application server N (608) receives the network packet via VNIC N (620), thereby completing the virtual wire (632). In one or more embodiments of the invention, the virtual wire (632) may also be used to transmit network traffic in the opposite direction, i.e., from application server N (608) to application server M (606).
[0042] Further, as discussed above, multiple virtual wires may be combined to form a "virtual network path." For example, Figure 6B shows virtual network path R (636), which flows from client O (628), through the router (602), through the firewall (604), and terminates at application server M (606). Specifically, the virtual network path R (636) includes the following virtual wires. A virtual wire is located between the physical network interface (626) and VNIC H (610). Another virtual wire is located between VNIC J (612) and VNIC K (614). Yet another virtual wire is located between VNIC L (616) and VNIC M (618). If the router (602) and the firewall (604) are located in the same blade, then a virtual switch may be substituted for the virtual wire located between VNIC J (612) and VNIC K (614), thereby eliminating use of the chassis interconnect (622) from communications between the router (602) and the firewall (604).
[0043] Similarly, Figure 6B shows virtual network path S (638), which flows from client P (630), through the router (602), and terminates at application server N (608). Virtual network path S (638) includes a virtual wire between the physical network interface (626) and VNIC H (610), and a virtual wire between VNIC J (612) and VNIC N (620). The differences between virtual network path R (636) and virtual network path S (638) exemplify how multiple virtual network paths may be located in the same blade chassis.
[0044] In one or more embodiments of the invention, VNIC settings are applied separately for each virtual network path. For example, different bandwidth limits may be used for virtual network path R (636) and virtual network path S (638). Thus, the virtual network paths may be thought of as including many of the same features as traditional network paths (e.g., using Ethernet cables), even though traditional network wires are not used within the blade chassis. However, traditional network wires may still be required outside the blade chassis, for example between the physical network interface (626) and client O (628) and/or client P (630).
[0045] Figure 6C shows a diagram of the virtual network topology (640) resulting from the use of the virtual network path R (636), virtual network path S (638), and virtual wire (632) shown in Figure 6B. The virtual network topology (640) allows the various components of the network (i.e., router (602), firewall (604), application server M (606), application server N (608), client O (628), and client P (630)) to interact in a manner similar to a traditional wired network. However, as discussed above, communication between the components located within the blade chassis (i.e., router (602), firewall (604), application server M (606), and application server N (608)) is accomplished without the use of traditional network wires.
[0046] In one embodiment of the invention, data may be transferred between virtual machines executing on different blades in a blade chassis using Transmission Control Protocol (TCP) and Internet Protocol (IP). Further, data may also be transferred between the virtual machines using low-overhead data transfers. In particular, data may be transferred directly from physical memory on one blade to physical memory on another blade.
[0047] More specifically, the virtual machine (or application executing therein) may establish a TCP connection with another virtual machine and then, using the TCP connection, perform a zero-copy handshake. In one embodiment of the invention, the zero-copy handshake involves determining whether the virtual machines are able to communicate using low-overhead data transfer and if the virtual machines (or applications executing therein) want to transfer data using low-overhead data transfer. In one embodiment of the invention, the virtual machines may communicate using a combination of data transfer over TCP/IP and data transfer using low-overhead data transfer.
[0048] In one embodiment of the invention, low-overhead data transfer is achieved by allowing the direct transfer of data from the virtual memory associated with a sending application (executing in a first virtual machine) to the virtual memory of a receiving application (executing in a second virtual machine), where the first application is executing on a first blade and the second application is executing on a second blade. In one embodiment of the invention, the target virtual memory address for the transfer must be provided prior to the transfer of data. If the receiving application is executing in a guest operating system (executing in a virtual machine), which in turn is executing in a host operating system, then the receiving application must provide the sending application (or a related process) a physical memory address (which corresponds to the virtual memory associated with the receiving application) for a buffer to which to transfer the data. However, the receiving application is only able to provide a virtual memory address for the receiving application. This virtual memory address must be translated one or more times in order to obtain the underlying physical memory address. The process of translation is described in Figure 7 below. Once the translation is complete, the physical memory address (as well as any other necessary information) is provided to the sending application (or a related process) to perform low-overhead data transfer as described in Figure 8.
[0049] Figure 7 shows a flowchart of a method for pre-posting buffers for an application prior to the application using low-overhead data transfer. In one or more embodiments of the invention, one or more of the steps shown in Figure 7 may be omitted, repeated, and/or performed in a different order than the order shown in Figure 7. Accordingly, embodiments of the invention should not be considered limited to the specific arrangement of steps shown in Figure 7.
[0050] In Step 700, an application specifies a pre-post buffer address. In one embodiment of the invention, the pre-post buffer address is a virtual memory address in virtual memory associated with the application. In one embodiment of the invention, the pre-post buffer address may refer to a buffer that is greater than 1 megabyte in size. In Step 702, the guest operating system receives and translates the pre-post buffer address into a guest OS virtual memory address. In one embodiment of the invention, the guest OS virtual memory address is a virtual memory address in a virtual memory associated with the guest operating system.
[0051] In Step 704, the guest operating system provides the guest OS virtual memory address to the host operating system. In Step 706, the host operating system receives and translates the guest OS virtual memory address into a host OS virtual memory address. Based on the host virtual memory address, the operating system may determine the underlying physical memory address corresponding to the host OS virtual memory address. The physical memory address corresponds to the host OS virtual memory address is the same physical memory address which corresponds to the pre-post buffer address.
[0052] In one embodiment of the invention, the host operating system notifies that the guest operating system that the per-post buffer address has been successfully pre-posted. The guest operating may, in turn, notify the application that the pre-post buffer address has been successfully pre-posted. In addition, the host operating system may maintain the translated physical address and any other related information (collectively referred to as "pre-post buffer information").
[0053] At this stage, the application may now participate in low-overhead data transfer. More specifically, the application may receive data using low-overhead data transfer. Those skilled in the art will appreciate the Figure 7 may be repeated multiple times for a given application in order for the application to pre-post multiple buffers for use in low-overhead data transfer. Further, the application may also send data to another application using low-overhead data transfer if the other application also pre-posts buffers using, for example, the method shown in Figure 7.
[0054] Figure 8 shows a flowchart of a method for initiating and using low-overhead data transfer. In one or more embodiments of the invention, one or more of the steps shown in Figure 8 may be omitted, repeated, and/or performed in a different order than the order shown in Figure 8. Accordingly, embodiments of the invention should not be considered limited to the specific arrangement of steps shown in Figure 8.
[0055] In Step 800, Application A attempts to initiate a TCP connection with Application B. In one embodiment of the invention, Application A provides an IP address assigned to the virtual machine (or assigned to the VNIC associated with the virtual machine) on which Application B is executing. In addition, Application A may also provide a port number.
[0056] In Step 802, the guest OS kernel, in response the request from Application A to initiate the TCP connection, creates socket A. In one embodiment of the invention, socket A is a kernel level process identified by the IP-Port Number pair and is a communication end-point configured to interface with Application A and the VNIC executing on the host operating system (on which the guest OS is executing). In Step 804, a TCP connection is initiated by socket A. In Step 806, socket B responds to the connection request and a TCP connection is established.
[0057] In Step 808, the zero-copy handshake is initiated. In one embodiment of the invention, the zero-copy handshake is an exchange of data designed to establish whether two applications may transfer data using low-overhead data transfer. In one embodiment of the invention, the zero-copy handshake is initiated when Application A sends one or more requests to Application B to determine whether Application A and Application B may transfer data using low-overhead data transfer. In one embodiment of the invention, the request may include placing a specific marker in a TCP SYN packet.
[0058] In one embodiment of the invention, instead of the applications initiating the zero-copy handshake, the VNICs executing on the respective host operating systems (see Figure 9 below) may initiate and subsequently perform the zero-copy handshake. In such cases, one or both of the applications, prior to the initiation of the TCP connection have indicated that they are able to transfer data using low-overhead data transfer and have performed the method shown in Figure 7 to obtain the pre-post buffer information.
[0059] In Step 810, as part of the zero-copy handshake, a determination is made about whether Application A and Application B are connected over a local TCP connection. In one embodiment of the invention, Application A and Application B are connected over a local TCP connection when both Application A and Application B are executing on blades within the same blade chassis. If Application A and Application B are connected over a local TCP connection, the process proceeds to Step 812. Alternatively, the process proceeds to Step 820. In Step 820, Applications A and B communicate using TCP/IP.
[0060] In Step 812, as part of the zero-copy handshake, a determination is made about whether Application B wants to participate in low-overhead data transfer. In one embodiment of the invention, this determination may include either of the following determinations: (i) Application B will send data to Application A using low-overhead data transfer but will only receive data from Application A via TCP/IP; and (ii) Application B will send data to Application A using low-overhead data transfer and Application B will receive data from Application A using low-overhead data transfer. If Application B wants to participate in low-overhead data transfer, then the process proceeds to Step 814. Alternatively, the process proceeds to Step 820 (i.e., Application B does not want to participate in either of the aforementioned scenarios). In one embodiment of the invention, the zero-copy handshake is performed over the TCP connection.
[0061] In Step 814, Application B is provided with Application A's pre-post buffer information. In Step 816, depending on the determination in Step 812, Application A may be provided with Application B's pre-post buffer information. In one embodiment of the invention, the information transferred in Step 814 and Step 816 are communicated over the TCP connection. In Step 818, Applications A and B participate in low-overhead data transfer.
[0062] In one or more embodiments of the invention, low-overhead data transfer from Application A to Application B uses, for example, a Direct Memory Access (DMA) operation, where the DMA operation uses as input Application B's pre-post buffer information. Those skilled in the art will appreciate that other write operations (e.g,. RDMA) may be used to write data directly from one physical memory location to another physical memory on different blades.
[0063] In one embodiment of the invention, the low-overhead transfer is performed by DMA (or RDMA) engines executing in (or managed by) the respective host operating systems. Further, because the data transfer is directly from the one blade to another, the data transfer does not require the additional processing overhead associated with other transfer protocols such as TCP. Further, in one embodiment of the invention, the low-overhead data transfer may use the underlying error detection and correction functionality of the chassis interconnect to ensure that data is transferred in an uncorrupted manner.
[0064] In one embodiment of the invention, once data from Application B is transferred to Application A using the low-overhead data transfer, Application A is notified of the presence of the data. In one embodiment of the invention, Application A receives the notification from the guest operating system on which it is executing. Further, the guest operating system is notified by the host operating system on which it is executing. Finally, the host operating system is notified by Application B, the guest operating system on which Application B is executing, or the host operating system on which the aforementioned guest operating system is executing (or a process executing thereon).
[0065] In one embodiment of the invention, Application A and Application B may communicate using both TCP/IP and low-overhead data transfer. For example, TCP/IP may be used for all communication of a certain type (e.g., all files in a specific file format) and/or less than a certain size and low-overhead data transfer may be used for all communication of another type and/or greater than a certain size.
[0066] Figure 9 shows an example of low-overhead data transfer in accordance with one or more embodiments of the invention. Figure 9 is provided for exemplary purposes only and should not be construed as limiting the scope of the invention. Referring to Figure 9, blade A (900) and blade B (902) are each communicatively coupled with a chassis interconnect (912). Application A (908) in blade A (900) is configured to communicate with application B (910) in blade B (902) via a TCP connection having socket A (918) and socket B (920) as endpoints. Specifically, socket A (918) is configured to transfer data to socket B (902) by way of VNIC A (926), VNIC B (928), and the chassis interconnect (912). Further, application A (908) is executing in virtual machine A (904) on guest OS A (not shown) and application B (910) is executing in virtual machine B (906) on guest OS B (not shown).
[0067] Based on the above, consider the scenario in which application A (908) and application B (910) each have performed the method described in Figure 7 to generate buffer pre-post information. More specifically, application A (908) allocated per-post buffer A (not shown) in Application A virtual memory (VM) (914). The virtual memory address associated with per-post buffer A is then translated to a guest operating system VM (922) address. The guest operating system VM (922) address is then translated by the host operating system A (930) to obtain a host VM address from the host VM (934), which corresponds to an underlying physical memory address. A similar process is performed for Application B (910) and using Application B VM (916) and translating to a guest operating system VM (924) address and finally to an underlying physical memory address which corresponds to the a host VM address in host VM (936)..
[0068] Using the above pre-post buffer information, the applications may communicate as follows in accordance with one embodiment of the invention. Specifically, application A (908) is configured to request a TCP connection with application B (910) for transferring data. Socket A (918) initiates a TCP connection with socket B (920) via VNIC A (926) to VNIC B (928).
[0069] Once the TCP connection is established, the zero-copy handshake is performed. Specifically, a determination is made by VNIC A (926) that Application A (908) and Application B (910) are connected over a local TCP connection. A further determination is made that Application B (910) will send data to Application A (908) using low-overhead data transfer and Application B (910) will receive data from Application A (908) using low-overhead data transfer.
[0070] In one or more embodiments of the invention, VNIC A (926) then passes Application A's pre-post buffer information to VNIC B (928) and VNIC B (928) passes Application B's pre-post buffer information to VNIC A (926). The applications may then transfer data using low-overhead data transfer.
[0071] In one embodiment of the invention, data from application B (910) is transferred using a RDMA engine and the application A's pre-post buffer information directly to applications A's VM (914), where the RDMA engine located on blade B (902) and is managed by VNIC B (928). Prior to the transfer, VNIC A may compare the location in the physical memory received from VNIC B with an allowed address range associated with application A to determine whether the data may be transferred to the location in memory specified by the pre-post buffer information. If the location in physical memory received by VNIC A is outside the allowed address range, then the transfer may be denied.
[0072] Embodiments of the invention may be also be used to transfer data applications by using embodiments of the invention to transfer data between virtual machines (e.g., virtual machine A (904) and virtual machine B (906)). For example, referring to Figure 9, to send data from application A (908) to application B (910). Application A (908) may transfer data over the connection to VNIC A (926). VNIC A (926) in accordance with embodiments of the invention, obtains pre-post buffer for virtual machine B (906) and subsequently transfers the data using, for example, a RDMA engine directly to the virtual Guest OS B VM (924). Upon receipt, the data is copied into application B VM (916). In such cases, the virtual machines, as opposed to the applications, are aware of the ability to transfer data using low-overhead data transfer. However, the applications are not aware of this functionality. Further, the applications, in this scenario, do not need to include functionality to pre-post buffers. Instead, the virtual machines need to include functionality to pre-post buffers.
[0073] Those skilled in the art will appreciate that while the invention has been described with respect to using blades, the invention may be extended for use with other computer systems, which are not blades. Specifically, the invention may be extended to any computer, which includes at least memory, a processor, and a mechanism to physically connect to and communicate over the chassis interconnect. Examples of such computers include, but are not limited to, multi-processor servers, network appliances, and light-weight computing devices (e.g., computers that only include memory, a processor, a mechanism to physically connect to and communicate over the chassis interconnect), and the necessary hardware to enable the aforementioned components to interact.
[0074] Further, those skilled in the art will appreciate that if one or more computers, which are not blades, are not used to implement the invention, then an appropriate chassis may be used in place of the blade chassis.
[0075] A computer program product comprising software instructions can perform embodiments of the invention. The software instructions may be stored on a computer readable medium such as a compact disc (CD), a diskette, a tape, or any other computer readable storage device.
[0076] While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of this disclosure, will appreciate that other embodiments can be devised which do not depart from the scope of the invention as disclosed herein. Accordingly, the scope of the invention should be limited only by the attached claims.
initiating (800), by a first application (908), a Transmission Communication Protocol, "TCP", connection with a second application (910), wherein the first application is executing on a first computer (900) in a first virtual machine (904), the second application is executing on a second computer (902) in a second virtual machine (906), and the first computer and the second computer are located on a chassis and communicate over a chassis interconnect (912), wherein the chassis interconnect comprises a Peripheral Component Interface Express (PCI-E) backplane and the first and second computers are configured to communicate with each other via PCI-E endpoints;
establishing (806), in response to the initiation, the TCP connection between the first application and the second application;
selecting (812), by a first virtual network interface card, "VNIC" (926), a second protocol from a group consisting of a first and the second protocol based on a determination (810) that the TCP connection between the first application and the second application is a local TCP connection and that the second application wants to communicate using the second protocol, wherein the first VNIC is located on the first computer and is interposed between the first virtual machine and the chassis interconnect, wherein the TCP connection is the local TCP connection when the first application and the second application are executing on separate physical computers connected to the chassis, and wherein the first protocol is TCP and the second protocol is a low-overhead data transfer; and
based on the selection of the second protocol,
providing (814), by the first application, pre-post buffer information to the second application, wherein the pre-post buffer information corresponds to a location in a physical memory of the first computer and wherein the location in physical memory corresponds to a virtual memory address of the first application, and
transferring (818) data, by the second application, to the first application using the pre-post buffer information, wherein transferring the data comprises writing the data directly into the location in the physical memory of the first computer.
generating the pre-post information, wherein generating the pre-post information comprises:
allocating (700) the virtual memory address in virtual memory associated with the first application;
providing (700) the virtual memory address to a guest operating system (OS) executing the first application, wherein the guest OS is executing in the first virtual machine;
translating (702) the virtual memory address to obtain a guest OS virtual memory address associated with the guest operating system;
providing (704) the guest OS virtual memory address to a host operating system upon which the guest operating system is executing;
translating (706) the virtual memory address to obtain a host OS virtual memory address associated with the host operating system, wherein the host OS virtual memory address corresponds to the location in the physical memory of the first computer.
writing, by the second VNIC, the data to the location in the physical memory of the first computer using remote direct memory access, "RDMA", and the location in the physical memory of the first computer.
a chassis interconnect (912); and
a first application (908) configured to execute in a first virtual machine (904) on a first computer (900) and a second application (910) configured to execute in a second virtual machine (906) on a second computer (902), wherein the first computer and the second computer are located on a chassis and communicate over the chassis interconnect,
wherein the chassis interconnect comprises a Peripheral Component Interface Express (PCI-E) backplane and the first and second computers are configured to communicate with each other via PCI-E endpoints,
wherein the first application is configured to initiate a Transmission Communication Protocol (TCP) connection with the second application, wherein, in response to the initiation, the TCP connection is established between the first application and the second application,
wherein a first virtual network interface card, "VNIC", executing on the first computer and interposed between the first virtual machine and the chassis interconnect, is configured to select a second protocol from a group consisting of a first protocol and the second protocol based on a determination that the TCP connection between the first application and the second application is a local TCP connection and that the second application wants to communicate using the second protocol, wherein the TCP connection is the local TCP connection when the first application and the second application are executing on separate physical computers connected to the chassis, and wherein the first protocol is TCP and the second protocol is a low-overhead data transfer,
wherein the first application is configured to provide pre-post buffer information to the second application in response to a determination that the first application is executing on the same chassis as the second application,
wherein the pre-post buffer information corresponds to a location in a physical memory of the first computer and wherein the location in physical memory corresponds to a virtual memory address of the first application, and
wherein the second application transfers data to the first application using the pre-post buffer information, wherein transferring the data comprises writing the data directly into the location in the physical memory of the first computer.
allocating the virtual memory address in virtual memory associated with the first application;
providing the virtual memory address to a guest operating system (OS) executing the first application, wherein the guest OS is executing in the first virtual machine;
translating the virtual memory address to obtain a guest OS virtual memory address associated with the guest operating system;
providing the guest OS virtual memory address to a host operating system upon which the guest operating system is executing;
translating the virtual memory address to obtain a host OS virtual memory address associated with the host operating system, wherein the host OS virtual memory address corresponds to the location in the physical memory of the first computer.
writing, by the VNIC, the data to the location in the physical memory of the first computer using remote direct memory access (RDMA) and the location in the physical memory of the first computer.
Einleiten (800) durch eine erste Anwendung (908) einer Übertragungskommunikationsprotokollverbindung, "TCP-Verbindung", mit einer zweiten Anwendung (910), wobei die erste Anwendung auf einem ersten Computer (900) in einer ersten virtuellen Maschine (904) läuft, wobei die zweite Anwendung auf einem zweiten Computer (902) in einer zweiten virtuellen Maschine (906) läuft und wobei sich der erste Computer und der zweite Computer auf einem Gestell befinden und über ein Gestellverbindungselement (912) kommunizieren, wobei das Gestellverbindungselement eine "Peripheral Component Interface Express"-Rückwandplatine (PCI-E-Rückwandplatine) umfasst und der erste und der zweite Computer konfiguriert sind, miteinander über PCI-E-Endpunkte zu kommunizieren;
Einrichten (806) der TCP-Verbindung zwischen der ersten Anwendung und der zweiten Anwendung als Reaktion auf die Einleitung;
Auswählen (812) durch eine erste virtuelle Netzschnittstellenkarte, "VNIC", (926) eines zweiten Protokolls aus einer Gruppe, die aus einem ersten und dem zweiten Protokoll besteht, anhand einer Bestimmung (810), dass die TCP-Verbindung zwischen der ersten Anwendung und der zweiten Anwendung eine lokale TCP-Verbindung ist und dass die zweite Anwendung unter Verwendung des zweiten Protokolls kommunizieren will, wobei sich die erste VNIC auf dem ersten Computer befindet und zwischen die erste virtuelle Maschine und das Gestellverbindungselement eingefügt ist, wobei die TCP-Verbindung die lokale TCP-Verbindung ist, wenn die erste Anwendung und die zweite Anwendung auf getrennten physikalischen Computern, die mit dem Gestell verbunden sind, laufen, und wobei das erste Protokoll TCP ist und das zweite Protokoll eine Datenübertragung mit geringem Overhead ist; und
anhand der Auswahl des zweiten Protokolls
Liefern (814) durch die erste Anwendung von Vor-Nach-Zwischenspeicherinformationen an die zweite Anwendung, wobei die Vor-Nach-Zwischenspeicherinformationen einem Platz in einem physikalischen Speicher des ersten Computers entsprechen und wobei der Platz im physikalischen Speicher einer virtuellen Speicheradresse der ersten Anwendung entspricht, und
Übertragen (818) von Daten durch die zweite Anwendung an die erste Anwendung unter Verwendung der Vor-Nach-Zwischenspeicherinformationen, wobei das Übertragen der Daten umfasst, die Daten unmittelbar an den Platz im physikalischen Speicher des ersten Computers zu schreiben.
Erzeugen der Vor-Nach-Informationen, wobei das Erzeugen der Vor-Nach-Informationen Folgendes umfasst:
Zuteilen (700) der virtuellen Speicheradresse im virtuellen Speicher, der der ersten Anwendung zugeordnet ist;
Liefern (700) der virtuellen Speicheradresse an ein Gastbetriebssystem (OS), das die erste Anwendung ausführt, wobei das Gast-OS in der ersten virtuellen Maschine läuft;
Übersetzen (702) der virtuellen Speicheradresse, um eine virtuelle Gast-OS-Speicheradresse, die dem Gastbetriebssystem zugeordnet ist, zu erhalten;
Liefern (704) der virtuellen Gast-OS-Speicheradresse an ein Host-Betriebssystem, auf dem das Gastbetriebssystem läuft;
Übersetzen (706) der virtuellen Speicheradresse, um eine virtuelle Host-OS-Speicheradresse zu erhalten, die dem virtuellen Host-Betriebssystem zugeordnet ist, wobei die virtuelle Host-OS-Speicheradresse dem Platz im physikalischen Speicher des ersten Computers entspricht.
Schreiben durch die zweite VNIC der Daten an den Platz im physikalischen Speicher des ersten Computers unter Verwendung eines entfernten Direktspeicherzugriffs, "RDMA", und des Platzes im physikalischen Speicher des ersten Computers.
ein Gestellverbindungselement (912); und
eine erste Anwendung (908), die konfiguriert ist, in einer ersten virtuellen Maschine (904) auf einem ersten Computer (900) zu laufen, und eine zweite Anwendung (906), die konfiguriert ist, in einer zweiten virtuellen Maschine (910) auf einem zweiten Computer (902) zu laufen, wobei sich der erste Computer und der zweite Computer auf einem Gestell befinden und über das Gestellverbindungselement kommunizieren,
wobei das Gestellverbindungselement eine "Peripheral Component Interface Express"-Rückwandplatine (PCI-E-Rückwandplatine) umfasst und der erste und der zweite Computer konfiguriert sind, miteinander über PCI-E-Endpunkte zu kommunizieren,
wobei die erste Anwendung konfiguriert ist, eine Übertragungskommunikationsprotokollverbindung (TCP-Verbindung) mit der zweiten Anwendung einzuleiten,
wobei als Reaktion auf die Einleitung die TCP-Verbindung zwischen der ersten Anwendung und der zweiten Anwendung eingerichtet wird,
wobei eine erste virtuelle Netzschnittstellenkarte, "VNIC", die auf dem ersten Computer läuft und zwischen die erste virtuelle Maschine und das Gestellverbindungselement eingefügt ist, konfiguriert ist, anhand einer Bestimmung, dass die TCP-Verbindung zwischen der ersten Anwendung und der zweiten Anwendung eine lokale TCP-Verbindung ist und dass die zweite Anwendung unter Verwendung des zweiten Protokolls kommunizieren will, ein zweites Protokoll aus einer Gruppe auszuwählen, die aus einem ersten Protokoll und dem zweiten Protokoll besteht, wobei die TCP-Verbindung die lokale TCP-Verbindung ist, wenn die erste Anwendung und die zweite Anwendung auf getrennten physikalischen Computern, die mit dem Gestell verbunden sind, laufen, und wobei das erste Protokoll TCP ist und das zweite Protokoll eine Datenübertragung mit geringem Overhead ist,
wobei die erste Anwendung konfiguriert ist, Vor-Nach-Zwischenspeicherinformationen als Reaktion auf eine Bestimmung, dass die erste Anwendung auf demselben Gestell wie die zweite Anwendung läuft, an die zweite Anwendung zu liefern,
wobei die Vor-Nach-Zwischenspeicherinformationen einem Platz in einem physikalischen Speicher des ersten Computers entsprechen und wobei der Platz im physikalischen Speicher einer virtuellen Speicheradresse der ersten Anwendung entspricht, und
wobei die zweite Anwendung unter Verwendung der Vor-Nach-Zwischenspeicherinformationen Daten an die erste Anwendung überträgt, wobei das Übertragen der Daten umfasst, die Daten unmittelbar an den Platz im physikalischen Speicher des ersten Computers zu schreiben.
Zuteilen der virtuellen Speicheradresse im virtuellem Speicher, der der ersten Anwendung zugeordnet ist;
Liefern der virtuellen Speicheradresse an ein Gast-Betriebssystem (Gast-OS), das die erste Anwendung ausführt, wobei das Gast-OS in der ersten virtuellen Maschine läuft;
Übersetzen der virtuellen Speicheradresse, um eine virtuelle Gast-OS-Speicheradresse, die dem Gastbetriebssystem zugeordnet ist, zu erhalten;
Liefern der virtuellen Gast-OS-Speicheradresse an ein Host-Betriebssystem, auf dem das Gastbetriebssystem läuft;
Übersetzen der virtuellen Speicheradresse, um eine virtuelle Host-OS-Speicheradresse zu erhalten, die dem Host-Betriebssystem zugeordnet ist, wobei die virtuelle Host-OS-Speicheradresse dem Platz im physikalischen Speicher des ersten Computers entspricht.
Schreiben durch die VNIC der Daten an den Platz im physikalischen Speicher des ersten Computers unter Verwendung eines entfernten Direktspeicherzugriffs (RDMA) und des Platzes im physikalischen Speicher des ersten Computers.
initialisation (800), par une première application (908), d'une connexion de protocole de communication en transmission, noté "TCP", avec une deuxième application (910), la première application s'exécutant sur un premier ordinateur (900) dans une première machine virtuelle (904), la deuxième application s'exécutant sur un deuxième ordinateur (902) dans une deuxième machine virtuelle (906) et le premier ordinateur et le deuxième ordinateur étant implantés sur un châssis et communiquant par le biais d'une interconnexion (912) de châssis, l'interconnexion de châssis comprenant un fond de panier d'interface de composants périphériques express (PCI-E) et les premier et deuxième ordinateurs étant configurés pour communiquer l'un avec l'autre via des points d'extrémité PCI-E ;
établissement (806), en réponse à l'initialisation, de la connexion TCP entre la première application et la deuxième application ;
sélection (812), par une première carte d'interface de réseau virtuel, notée "VNIC", (926), d'un deuxième protocole dans le groupe constitué par un premier protocole et le deuxième protocole sur la base d'une détermination (810) que la connexion TCP entre la première application et la deuxième application est une connexion TCP locale et que la deuxième application souhaite communiquer au moyen du deuxième protocole, la première VNIC étant implantée sur le premier ordinateur et étant intercalée entre la première machine virtuelle et l'interconnexion de châssis, la connexion TCP étant la connexion TCP locale lorsque la première application et la deuxième application s'exécutent sur des ordinateurs physiques distincts connectés au châssis, et le premier protocole étant le TCP et le deuxième protocole étant un transfert de données à faible surdébit ; et
sur la base de la sélection du deuxième protocole,
fourniture (814), par la première application, d'informations de tampon de pré-affectation à la deuxième application, les informations de tampon de pré-affectation correspondant à un emplacement dans une mémoire physique du premier ordinateur et l'emplacement dans la mémoire physique correspondant à une adresse de mémoire virtuelle de la première application, et
transfert (818) de données, par la deuxième application, à la première application au moyen des informations de tampon de pré-affectation, le transfert des données comprenant l'écriture des données directement à l'emplacement dans la mémoire physique du premier ordinateur.
génération des informations de pré-affectation, la génération des informations de pré-affectation comprenant les étapes suivantes :
allocation (700) de l'adresse de mémoire virtuelle dans une mémoire virtuelle associée à la première application ;
fourniture (700) de l'adresse de mémoire virtuelle à un système d'exploitation (OS) invité exécutant la première application, l'OS invité s'exécutant dans la première machine virtuelle ;
traduction (702) de l'adresse de mémoire virtuelle afin d'obtenir une adresse de mémoire virtuelle d'OS invité associée au système d'exploitation invité ;
fourniture (704) de l'adresse de mémoire virtuelle d'OS invité à un système d'exploitation hôte sur lequel le système d'exploitation invité s'exécute ;
traduction (706) de l'adresse de mémoire virtuelle afin d'obtenir une adresse de mémoire virtuelle d'OS hôte associée au système d'exploitation hôte, l'adresse de mémoire virtuelle d'OS hôte correspondant à l'emplacement dans la mémoire physique du premier ordinateur.
écriture, par la deuxième VNIC, des données à l'emplacement dans la mémoire physique du premier ordinateur au moyen d'un accès distant direct en mémoire, noté "RDMA", et de l'emplacement dans la mémoire physique du premier ordinateur.
une interconnexion (912) de châssis ; et
une première application (908) configurée pour s'exécuter dans une première machine virtuelle (904) sur un premier ordinateur (900) et une deuxième application (910) configurée pour s'exécuter dans une deuxième machine virtuelle (906) sur un deuxième ordinateur (902), le premier ordinateur et le deuxième ordinateur étant implantés sur un châssis et communiquant par le biais de l'interconnexion de châssis,
l'interconnexion de châssis comprenant un fond de panier d'interface de composants périphériques express (PCI-E) et les premier et deuxième ordinateurs étant configurés pour communiquer l'un avec l'autre via des points d'extrémité PCI-E,
la première application étant configurée pour initialiser une connexion de protocole de communication en transmission, noté "TCP", avec la deuxième application,
en réponse à l'initialisation, la connexion TCP étant établie entre la première application et la deuxième application,
une première carte d'interface de réseau virtuel, notée "VNIC", s'exécutant sur le premier ordinateur et intercalée entre la première machine virtuelle et l'interconnexion de châssis, étant configurée pour sélectionner un deuxième protocole dans le groupe constitué par un premier protocole et le deuxième protocole sur la base d'une détermination que la connexion TCP entre la première application et la deuxième application est une connexion TCP locale et que la deuxième application souhaite communiquer au moyen du deuxième protocole, la connexion TCP étant la connexion TCP locale lorsque la première application et la deuxième application s'exécutent sur des ordinateurs physiques distincts connectés au châssis, et le premier protocole étant le TCP et le deuxième protocole étant un transfert de données à faible surdébit,
la première application étant configurée pour fournir des informations de tampon de pré-affectation à la deuxième application en réponse à une détermination que la première application s'exécute sur le même châssis que celui de la deuxième application,
les informations de tampon de pré-affectation correspondant à un emplacement dans une mémoire physique du premier ordinateur et l'emplacement dans la mémoire physique correspondant à une adresse de mémoire virtuelle de la première application, et
la deuxième application transférant des données à la première application au moyen des informations de tampon de pré-affectation, le transfert des données comprenant l'écriture des données directement à l'emplacement dans la mémoire physique du premier ordinateur.
allocation de l'adresse de mémoire virtuelle dans une mémoire virtuelle associée à la première application ;
fourniture de l'adresse de mémoire virtuelle à un système d'exploitation (OS) invité exécutant la première application, l'OS invité s'exécutant dans la première machine virtuelle ;
traduction de l'adresse de mémoire virtuelle afin d'obtenir une adresse de mémoire virtuelle d'OS invité associée au système d'exploitation invité ;
fourniture de l'adresse de mémoire virtuelle d'OS invité à un système d'exploitation hôte sur lequel le système d'exploitation invité s'exécute ;
traduction de l'adresse de mémoire virtuelle afin d'obtenir une adresse de mémoire virtuelle d'OS hôte associée au système d'exploitation hôte, l'adresse de mémoire virtuelle d'OS hôte correspondant à l'emplacement dans la mémoire physique du premier ordinateur.
l'écriture, par la VNIC, des données à l'emplacement dans la mémoire physique du premier ordinateur au moyen d'un accès distant direct en mémoire (RDMA) et de l'emplacement dans la mémoire physique du premier ordinateur.
REFERENCES CITED IN THE DESCRIPTION
This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.
Patent documents cited in the description
Non-patent literature cited in the description
- D GOLDENBERG et al.Zero Copy Sockets Direct Protocol over InfiniBand - Preliminary Implementation and Performance AnalysisProceedings of the 13th Symposium on High Performance Interconnects, 2005, [0005]